Tropical forest had undergone rapid loss in the early 21st century, bringing about warming effects on regional climate mainly via changing evapotranspiration. Regional climate models are powerful tools in assessing the biophysical feedbacks of deforestation. As the key part of climate models, land surface schemes regulate the fluxes of heat and water vapour between land and atmosphere, which may largely affect the pattern and magnitude of how forest cover change impacts climate. Here we investigate the deforestation-induced warming effect in Southeast Asian Massif by employing three land surface schemes (NoahMP, CLM, and Noah mosaic) in the Weather Research and Forecasting (WRF) model. We perform the analyses by designing two sets of experiments with comparative land-cover scenarios derived from high-resolution forest cover change dataset during the local dry season. Model validations show that, regarding the magnitude, the CLM scheme is in good agreement with observed surface air temperature while the Noah mosaic scheme has the largest negative biases. When comparing the results between the two scenarios, it is found that only the Noah mosaic scheme which takes the sub-grid approach reasonably reproduces the response of warming effect to deforestation. By contrast, the NoahMP scheme fails to accurately capture the deforestation-induced regional warming due to the use of the dominant approach at grid level. The CLM, a scheme that theoretically considers all land cover types within girds and thus should have the capability to capture the climate feedbacks of deforestation, turns out to be less sensitive to forest loss in those grids where the dominant type remains unchanged, and presents the similar pattern of temperature change as the NoahMP scheme. Based on these results, we speculate that the CLM scheme takes the dominant approach instead of the all-type mosaic way when coupled into the WRF model. This work demonstrates that the Noah mosaic scheme could be temporally applied in simulating the climate feedbacks of land cover conversion. We suggest that the representations of sub-grid characteristics in the CLM scheme should be modified in the following version of the WRF model.